Suicide is a serious global public health problem. According to the World Health Organization, 726,000 people take their own lives each year, and many attempt suicides. All cases are tragedies for families and leave lasting effects for relatives. The study detected suicide risk in social media posts using Machine learning and Deep learning techniques. The proposed methodology comprises five phases: Acquisition of the dataset; Preprocessing (removal of useless information, character substitution, data augmentation, tokenization and lemmatization); Feature extraction (Word2Vec, GloVe and FastText); Implementation of Machine Learning models (LightGBM, CatBoost, SVM, Random Forest and Logistic Regression), Deep Learning (BERT, XLNet and ELECTRA); and Evaluation of the models. The results demonstrated the superior performance of the hybrid BERT+LSTM model with 96.52% Accuracy, 96.17% Precision, 96.50% F1-score, 96.84% Recall and 96.52% ROC-AUC. In conclusion, the results demonstrate that the hybrid model BERT+LSTM can optimally classify the risk of suicide present in a social network post, providing an efficient tool that allows early detection of suicide and timely treatment.